Isaacus models are now up to 2.6× faster

As part of our ongoing commitment to delivering the fastest and most reliable legal AI, Isaacus has shipped a series of backend improvements that make our models up to 2.6× faster.

Before-and-after request latency for Isaacus models across enrichment, embedding, reranking, classification, and extraction, showing speed-ups of up to 2.6×. — Median latency of an 8,192-token request, before and after our infrastructure and inference engine overhaul.

This speed-up is the result of a ground-up overhaul of our infrastructure and inference engine, spanning everything from eliminating network hops between our services to rewriting low-level kernels.

As a result, customers can expect lower request latencies, greater throughput, and improved stability across all Isaacus models.

We’ve also launched a new status page, where customers can subscribe to updates on maintenance windows, service disruptions, and performance improvements.

This is only the beginning. As we work toward the release of the Blackstone Graph, we’ll be rolling out further enhancements to the Isaacus platform yet to be announced.

We thank our customers for their patience as we worked behind the scenes to improve our models.

To stay up to date on our progress, follow us on LinkedIn, X/Twitter, or Reddit.